8000 How to handle tags · Issue #189 · eole-nlp/eole · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

How to handle tags #189

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Closed
chillum-codeX opened this issue Jan 20, 2025 · 6 comments
Closed

How to handle tags #189

chillum-codeX opened this issue Jan 20, 2025 · 6 comments

Comments

@chillum-codeX
Copy link
chillum-codeX commented Jan 20, 2025 •
edited by francoishernandez
Loading

I am training a model for en to spanish and spanish to english but when I am doing translation unable to handle tags
Input
for sutures is an art<g1> </g1>learned through practice, the knowledge of what<g2> </g2>happens at the cellular and

The output should looks like
Les sutures sont un art<g1> </g1> appris par la pratique, la connaissance de ce qui<g2> </g2> se passe au niveau cellulaire.

But I am getting
Translation:
Les coutures sont un art appris par la pratique, la connaissance de ce qui se passe au niveau cellulaire et pour
@vince62s
Copy link
Contributor

What is probably super clear in your head is not in your post above, please clarify what you mean with your tag thing and post your config.

@chillum-codeX
Copy link
Author
chillum-codeX commented Jan 20, 2025 •
edited by francoishernandez
Loading

Input
for sutures is an art<g1> </g1>learned through practice, the knowledge of what<g2> </g2>happens at the cellular and

Output
fra_Latn
Les coutures sont un art appris par la pratique, la connaissance de ce qui se passe au niveau cellulaire et pour

The translation is not maintaining the tags

I am using pretrained weight nllb-200-1.3B-onmt.pt

Or any suggestions would help me to fine-tune the data on pretrained weight
I need to handle tags also while translating.

@chillum-codeX
Copy link
Author

Any suggestions will help me a lot.

@francoishernandez
Copy link
Member

@chillum-codeX I edited your comments with backticks to properly show the html-like tags, otherwise hidden by Github rendering.

Most models should be able to naively output some form of tags in examples such as provided, even if not placed properly. The fact that there are none at all might indicate some tokenization/vocab related issue. (E.g. if the tags are tokenized improperly, and can't be handled as is by the token.)

Also, you mention some pretrained model nllb-200-1.3B-onmt.pt. The .pt model format support has been dropped for quite a while, so you might want to start by migrating to a more recent codebase and models. If such conversion has already been done, provide details that might help understand the root cause of the issue.

@chillum-codeX
Copy link
Author

Than how to fine tune nllb with EOLE
I want to train en to spa and spa to eng on domain-specific data

Previously I have used openmt py

@francoishernandez
Copy link
Member
francoishernandez commented Jan 30, 2025 •
edited
Loading

I just opened #204 to facilitate the conversion of NLLB models. I'll close this issue as it's not very specific. If you need further assistance, please open a topic in the Discussions tab, which is more appropriate for broad support requests.

I probably won't add any finetuning recipe there for now, but it should be quite straightforward to adapt from the opennmt ones (#69). Feel free to PR a valid finetuning config once you get it running!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
3 participants
@francoishernandez @vince62s @chillum-codeX
0